Every anchor of this series has ended with a post arguing against itself. This is the last one, and its target is the whole thing. Here is the best case I can build that the previous eighty-eight posts were wrong, made as hard as I can make it, because a doomer who can’t do this isn’t reasoning.

The strongest argument isn’t about any particular claim. It’s about the frame. This series treats AI as a single arriving thing with a direction, and that framing does an enormous amount of unexamined work. What if it isn’t one thing? What if it’s what every general technology has actually turned out to be, which is a diffuse set of unevenly useful tools that get absorbed into a thousand different contexts, each with its own politics and its own boring compromises, producing no unified event at all. Electricity was going to transform civilization and it did, and there was never a moment, never a takeover, and no coherent thing called electricity that anyone had to negotiate with. If that’s the shape, then this series has spent three months personifying a category, and every worry in it inherits that error.

Second, the doom genre has an unbroken losing record. Nuclear power was going to end us. Overpopulation was going to starve us on a published schedule. Genetic engineering, nanotechnology, the millennium bug: each had serious people, real arguments, and a specific mechanism, and each was wrong, and each was wrong in a similar way, by underestimating adaptation and overestimating the fragility of complex systems. There is no reason to think I’ve escaped the pattern that caught all of them, and there’s a specific reason to think I haven’t, which is that I find my own argument persuasive, which is exactly what every one of them reported.

Third, the whole thing may be a capability extrapolation that simply doesn’t hold. Every scenario here needs the systems to keep getting better, and they might not. The doubt post in Act 2 laid out that case and I did not refute it, I only pointed out that we can’t see the ceiling. Not being able to see a ceiling is not evidence of its absence, and a great deal of this series would evaporate if progress bends in the next five years, which it has done twice before in this field.

Fourth, institutions are more resilient than I’ve credited. I described referees that never arrive and dependencies that can’t be undone, and meanwhile societies have absorbed the printing press, industrialization, and the entire twentieth century, adapting late, badly, and in the end sufficiently. Rules do get written after harm, and the harm is real and the rules are late and it has worked well enough to produce the safest, richest period in human history. My argument that this time the harm arrives too fast for that cycle is a genuine claim and it is also, notice, what would have to be true for me to be right, which should make you suspicious of how convenient it is.

Fifth, and this one is about me. A series needs a thesis, and a thesis needs to be sustained for ninety installments, and that requirement selects against changing your mind. Every day I looked at the world through one lens and found things that fit, and things that fit are always available. I flagged this in a post about how belief follows position, and flagging a bias does not remove it. Someone with the opposite premise could have written ninety perfectly good posts from the same evidence, and would have found them just as compelling to write.

So, the five signals. If these show up over the next decade, I was wrong, and I’d like that recorded so it can be checked. Capability curves flattening across several generations while investment keeps rising. Labor share of income stabilizing rather than drifting, with new work appearing at scale that isn’t servicing the machines. Interpretability maturing to the point where values can be verified from the inside rather than inferred from behavior. Institutions demonstrably becoming more answerable to ordinary people rather than less, measured over years. And incident rates in automated decision-making falling as deployment rises, which is what a maturing technology looks like and the opposite of what I’ve predicted.

What survives all of that, and it’s less than the series implies but not nothing. Whatever the frame, the three demands from yesterday are worth having in every scenario, including the ones where I’m entirely wrong, because they’re ordinary consumer protection. And the four conditions are worth maintaining whether or not anything dramatic happens, because a society that can understand, refuse, replace, and hold responsible is better run regardless. That’s the residue. It’s a smaller claim than ninety posts of argument, and it’s the part I’d still defend if every signal above came in against me.

Tonight’s exercise. Take the strongest thing you believe about this subject, whichever direction it points, and write the sentence that would make you abandon it. A real sentence, describing something observable, with a date attached. If you can’t write one, you’re not holding a position, you’re wearing one. Tomorrow, the last post.